A set of 148 chromosome segment substitution lines (CSSLs) was derived from the inter-subspecific cross between two genome-sequenced rice cultivars, japonica Nipponbare (NIP, as the donor) and indica Zhenshan97 (ZS97, as the recurrent parent), using a backcross scheme with a marker-assisted selection approach (Sun et al., 2015; Zhang et al., 2020b). A CSSL that harbors two loci/regions of interest was selected to cross with ZS97 to develop an F₁ hybrid. Deviation of the allele and genotype frequencies from Mendelian expectations at a target locus was assessed in segregating the progeny of the hybrid. Furthermore, from the CSSL-derived progenies, several independent near-isognice lines (NILs), each containing only a single introduced heterozygous segment covering either TRD4.1 or TRD4.2 in the otherwise uniform background of ZS97, were developed. Six independent segregating populations were derived from relevant NILs to validate TRD loci. To delimit TRD4.1 and TRD4.2, two large segregated populations were generated to select recombinant individuals per locus. The recombinants with heterozygous target regions were self-crossed to generate progeny for phenotyping TRD. In addition, the reported BIL population from the cross of NIP and 9,311, which were genotyped by the genotyping-by-sequencing method (Yuan et al., 2019), was also used for the analysis of epistatic interaction of particular TRD loci.

For reciprocal cross to test gametic selection, a line (named as NZ) carrying TRD4.1 at a heterozygous state was first obtained by crossing ZS97 with a NIL that harbors the NIP alleles at TRD4.1. Then, two reciprocal populations were generated by crossing NZ as male or female parent with a developed NIL (named as MM) that contains introduced Minghui 63 (MH) alleles at TRD4.1 (Chen et al., 2018). Both NZ and MM have the same genetic background as ZS97.

Uniformly germinated seeds were planted on 96-well plates with the bottoms removed (Li et al., 2015), and the plates were placed in a growth chamber (Dongnan, Ningbo, China) or a greenhouse at 30°C under 16-h light/8-h dark conditions. The 7-day seedlings were used for genotype analysis. Some lines and segregating populations were grown in the experimental field of Huazhong Agricultural University (HAU) in Wuhan (30.48N, 114.2E), China. Each line was planted in a row with 10 individual spacings of 16.7 cm × 26.6 cm for genotyping. The field management was managed according to the local standard practices.

Genomic DNA was extracted from young seedling leaves, as described previously (Zhang et al., 2020b). The genome-wide genotyping of CSSLs was conducted using the RICE 6K array (Sun et al., 2015; Chen et al., 2018). For the genotyping of segregating populations, a number of polymorphic markers, including simple sequence repeat and insertion/deletion markers, were designed (Supplementary Table 1) and used for PCR reaction following the described procedure (Panaud et al., 1996).

The allele and genotype frequencies were assayed by several polymorphic markers in a segregated population. The statistical Chi-square test (χ²) was performed using chisq.test function of R¹ to determine whether the observed frequencies are distorted from Mendelian segregation ratios in any segregated population. A genomic region with two or more distorted consecutive markers was considered as one with TRD. Also, the TRD analysis in two subsets of the BIL population was performed using a single-marker analysis model in QTL IciMapping (v4.1), as described previously (Zhang et al., 2020a). A significance level of LOD > 5.0 was set as the threshold to declare the presence of a putative TRD effect in a given bin/marker.

To determine the cause of TRD (e.g., gametic or zygotic selection) at a given locus, the allele and genotype (NIP and ZS97) frequencies were assayed to test if they fit the Mendelian ratios, respectively. The comparative patterns of the allele and genotype frequencies may infer gametic or zygotic selection in multiple F₂ populations. If the allele frequency does fit the theoretical ratio of 1:1, and the genotype frequency is biased to 1: 2: 1, the TRD may be raised by zygotic factors. In this regard, an additional test is also performed to determine whether the observed heterozygote frequency deviated from the theoretical genotype frequency (0.5) in the F₂ populations (Fishman and McIntosh, 2019). If the allele frequency is distorted from 1:1, and the genotype frequency fits the ratio of 1: 2: 1, the TRD may result from gametic factors. If there is a distortion of both allele and genotype frequencies from the Mendelian ratios, then the TRD is due to both gametic and zygotic factors.

To track male- and female-specific transmission patterns for a given region using a reciprocal crossing test, diagnostic markers that were polymorphic among the parental lines (NIP, ZS97, and MH) were used for genotyping in the reciprocal cross progeny. The pattern of unequal frequencies through gametic selection is illustrated in Figure 1. If the F₁ (ZN) is the male parent (or conversely the female), distortion of genotype (NM and ZM) frequencies in the progeny indicates the gametic selection at the F₁-heterozygous locus.

Pollen fertility was examined with the I₂-KI staining method using mixed pollen grains of more than eight florets from the panicles of each plant, as previously described (Li et al., 2017). A microscope field with at least 200 pollen grains was observed. Pollen fertility was scored as the percentage of filled and stained grains in total grains (Kubo et al., 2011; Li et al., 2017).

To detect the TRD effect of the introduced NIP segments on chromosome 4, one of the CSSLs that contained the particular segments was crossed with ZS97 to produce a segregating population (name CSSL91-derived population). Graphical genotype analysis showed that CSSL91 carried four introduced NIP segments within the ZS97 background, two on chromosome 4 being targeted, and two on chromosome 2 (Supplementary Figure 1). Initially, the population (n = 191) was genotyped using four polymorphic markers, each located in the corresponding segment. The allele and genotype frequencies at the two markers (M1 and M2) on chromosome 2 showed normal-Mendelian segregation ratios. However, a non-Mendelian segregation of the two markers (M3 and M4) on chromosome 4 was observed in both allele and genotype frequencies (Table 1 and Supplementary Figure 1). Furthermore, eight additional consecutive markers (M5 to M12) distributed in the two TRD regions were all distorted significantly from the expected Mendelian ratios in the allele and genotype frequencies. The distortion of these markers was all biased toward ZS97 (Table 1 and Supplementary Table 2), suggesting that the ZS97 alleles at these two regions/loci (named TRD4.1 and TRD4.2) were transmitted to the progeny at a higher frequency than the NIP alleles.

To validate the effect of TRD4.1, two lines that carried heterozygous TRD4.1 along with TRD4.2 homozygous ZS97 or NIP were obtained and self-crossed to produce corresponding segregating populations. As a cluster of molecular markers showing TRD suggests that the chromosomal region may have one or more genes causing TRD, the representative marker (M7) was used to analyze the TRD effect at TRD4.1. A non-Mendelian segregation of TRD4.1 was observed in these two populations (Figure 2 and Table 2). Moreover, the ZS97 alleles at TRD4.1 were preferentially transmitted to the progeny in heterozygotes.

To check the effect of TRD4.2, two lines that harbored heterozygous TRD4.2 and homozygous at TRD4.1 either with ZS97 or NIP alleles were self-pollinated to produce segregating populations (Figure 2 and Table 2). Consistently, TRD4.2 assayed by the representative marker M12 exhibited a significant TRD effect with the preferential transmission of the ZS97 alleles in the progeny of heterozygotes harboring homozygous TRD4.1^NIP. In contrast, normal segregation of TRD4.2 was observed in the progeny of the heterozygotes when carried with homozygous TRD4.1^ZS97 (Table 2).

An additional segregated population (n = 902) was also generated from the heterozygotes at both TRD4.1 and TRD4.2 (Figure 2). Frequencies of the nine genotypes assayed by two markers (M7 and M12) linked with TRD4.1 and TRD4.2 did not fit the Mendelian segregation ratio (χ² = 185.5; P < 2.2 × 10^–16), given the two loci are linked (Supplementary Table 2). In particular, the number of genotype homozygous TRD4.1^ZS97TRD4.2^ZS97 was more than that of TRD4.1^NIPTRD4.2^NIP, TRD4.1^ZS97TRD4.2^NIP, and TRD4.1^NIPTRD4.2^ZS97 in the population. Collective data confirmed that TRD arose from both TRD4.1 and TRD4.2 and a preferential transmission of the ZS97 gametes at the two loci.

To determine whether TRD4.1 affects other genomic regions on TRD, the BIL population that was previously developed from the cross of NIP and 9,311 (Zhang et al., 2020b) was divided into two subpopulations according to the TRD4.1 genotype: SubN (n = 75) in which all lines had TRD4.1^NIP/NIP and SubJ (n = 325) that had TRD4.1^9311/9311 assayed by 10 consecutive bin-markers (from 13.47 to 15.87 Mb) at the TRD4.1 region (Supplementary Table 3). Four and ten regions were detected in SubN and SubJ, respectively. Among them, three TRD regions (TR1.3, TR8.2, and TR12.1) were common in both subpopulations. The other seven regions were identified only in SubJ. In particular, the TRD4.2 region was detected only in SubN but not in SubJ (Supplementary Table 3). These results revealed the effect of TRD4.1 on the detection of other TRD loci in the BIL population.

To investigate whether TRD4.1 and TRD4.2 are involved in the gametic or zygotic factors, the population segregated at both TRD4.1 and TRD4.2 was analyzed. The results showed that the frequencies of heterozygous genotypes at either TRD4.1 or TRD4.2 were approximately 0.5. The observed frequencies of heterozygotes at either TRD4.1 or TRD4.2 exhibited no significant derivation from the expected frequencies (Supplementary Table 2). This non-distortion of heterozygote frequency may indicate a gametophytic barrier. In addition, normal pollen fertility (greater than 95%) of the nine genotypes revealed that TRD was not caused by pollen fertility (Supplementary Table 4). These results confirm that the gametic factors are involved in TRD4.1 and TRD4.2.

To investigate whether the TRD4.1 effect is caused by female or male gametic factors, a reciprocal test cross was made between NILs with a heterozygous (ZN) segment and a MM segment that contains homozygous Minghui 63 alleles at the locus (Figure 1). Segregation of the locus was assessed in a F₁ progeny from the crosses. The progenies of the reciprocal crosses (NZ × MM and MM × ZN) were examined for TRD effects. The polymorphic markers M13 and M14 that were tightly linked with TRD4.1 and which could distinguish the three genotypes (NIP, ZS97, and MH63) were used to classify two progeny genotypes (NM and ZM) of the crosses. The genotype frequency of TRD4.1 showed normal-Mendelian segregation (no TRD) by the Chi-square test against the expected ratio of 1:1 (NM:ZM = 283:292; χ² = 0.14; P > 0.70) in the cross of ZN × MM (n = 575), in which ZN was used as the female parent. The results suggest that NIP and ZS97 alleles were equally transmitted to the progeny. However, significant distortion of genotype frequencies (NM:ZM = 196:343; χ² = 40.09; P < 2.5 × 10^–10) was observed in the progeny of the cross of MM × NZ (n = 539), in which ZN was used as the male parent (Figure 1). In addition, normal pollen fertility (greater than 95%) of ZN was observed. These results indicate that the ZS97 allele was preferentially transmitted to the progeny over the NIP allele. Therefore, the distortion of genotype frequencies at TRD4.1 in the progeny involved male gamete selection.

To narrow the region of TRD4.1, a mapping population comprising of approximately 6,500 individuals was developed from one plant that harbored only one heterozygous TRD4.1 region and one homozygous TRD4.2^NIP within the ZS97 background. Initially, five recombinant individuals (R1 to R5) were obtained using several markers linked with TRD4.1. Then, the recombinants were self-crossed to generate five independent segregating populations for further genotyping (Figure 3). Based on frequencies of the genotypes in each recombinant-derived population using representative markers (M7 or M8), the distorted region covering TRD4.1 was determined in the corresponding recombinant. TRD4.1 was preliminarily mapped to an approximately 590-kb region between markers M7 and M8. For fine mapping of TRD4.1, eight recombinant individuals (R6 to R13) within the interval of M7-M8 were selected and generated eight independent recombinant-derived populations. TRD analysis based on frequency genotypes in recombinant-derived populations delimited TRD4.1 into a 100-kb region between markers M19 and M8.

Based on the same approach, fine-mapping of TRD4.2 was conducted on a segregated population (composed of approximately 5,800 individuals) derived from a plant that carried only one heterozygous TRD4.2 region and one homozygous TRD4.1^NIP segment in the ZS97 background. Seventeen recombinant individuals in the TRD4.2 region flanked by markers M23 and M12 were obtained (Figures 4A,B). Based on TRD analysis of genotype frequencies in each of the 17 recombination-derived populations and genotyping with additional markers in the target region, TRD4.2 was narrowed down to an approximately 34.1-kb region. This region contains nine predicted genes based on the reference genome² (Figure 4C). Of them, LOC_Os04g33150 was the only one expressed gene, after removing those annotated as unknown, transposons/retrotransposons, or hypothetical proteins. It was specifically and highly expressed in the pre-emergence inflorescence, 5-d seed, and 25-d endosperm.³ This gene encodes a desiccation-related protein. Sequence differences that existed in the coding region between NIP and ZS97 may cause two amino acid changes.⁴ Therefore, LOC_Os04g33150 is the most likely candidate gene for TRD4.2.